Xingjian Bai · 白行健

Ph.D. Student · MIT CSAIL

Hello! I am a second-year Ph.D. student at MIT, working with Prof. Kaiming He. I received my master's and bachelor's degrees in Mathematics and Computer Science from the University of Oxford. My research focuses on understanding and empirically building stronger generative models — recently, flows, diffusion, and video world models. Over the years, I have also explored theoretical computer science, with a focus on online and learning-augmented algorithms. More broadly, I am drawn to research that is both scientifically impactful and intellectually stimulating.

During my time at Oxford, I had the privilege of working with Prof. Jan Obloj, Prof. Christian Coester, Prof. Christian Rupprecht, and Luke Melas-Kyriazi. I also spent a rewarding summer at Stanford, collaborating with Prof. Jiajun Wu, Dr. Weiyu Liu, and Joy Hsu. Outside of my academic pursuits, I enjoy jogging and playing tennis. I also engage in algorithmic programming competitions.

Collaboration. If you would like to discuss anything related to research, startups, or life in general, I would be eager to hear from you. Please feel free to contact me via email or schedule a chat via Chatting Time.

Email GitHub Scholar X / Twitter CV Book a chat

News

Mar 2026 Check out our new preprint, UNITE, on single-stage joint training of tokenization and latent diffusion.
Feb 2026 Our paper, Causality in Video Diffusers is Separable from Denoising, is accepted at CVPR 2026!
May 2025 Our paper, Mean Flows for One-step Generative Modeling, is accepted at NeurIPS 2025 as an oral presentation.
Oct 2024 Our paper, Unweighted Layered Graph Traversal, is accepted at SODA 2025!
Jun 2024 My master's thesis, Efficient and Adaptive Image Generation with Fixed Point Diffusion Models, has won the Departmental Prize for the Best Computer Science Project.
Feb 2024 Our paper, Fixed Point Diffusion Models, is accepted at CVPR 2024!

Publications

End-to-End Training for Unified Tokenization and Latent Denoising (UNITE)

Shivam Duggal*, Xingjian Bai*, Zongze Wu, Richard Zhang, Eli Shechtman, Antonio Torralba, Phillip Isola, William T. Freeman

* equal contribution

arXiv 2026 MIT · Adobe Research

page arXiv code X

@article{duggal2026unite,
  title   = {End-to-End Training for Unified Tokenization and Latent Denoising},
  author  = {Shivam Duggal and Xingjian Bai and Zongze Wu and Richard Zhang and Eli Shechtman and Antonio Torralba and Phillip Isola and William T. Freeman},
  journal = {arXiv preprint arXiv:2603.22283},
  year    = {2026}
}

Causality in Video Diffusers is Separable from Denoising

Xingjian Bai, Guande He, Zhengqi Li, Eli Shechtman, Xun Huang, Zongze Wu

CVPR 2026 IEEE/CVF Conference on Computer Vision and Pattern Recognition

page arXiv code X

@inproceedings{bai2026scd,
  title     = {Causality in Video Diffusers is Separable from Denoising},
  author    = {Bai, Xingjian and He, Guande and Li, Zhengqi and Shechtman, Eli and Huang, Xun and Wu, Zongze},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2026}
}

Mean Flows for One-step Generative Modeling

Zhengyang Geng, Mingyang Deng, Xingjian Bai, J. Zico Kolter, Kaiming He

NeurIPS 2025 · Oral CMU · MIT

oral arXiv code X

@inproceedings{geng2025meanflow,
  title     = {Mean Flows for One-step Generative Modeling},
  author    = {Geng, Zhengyang and Deng, Mingyang and Bai, Xingjian and Kolter, J. Zico and He, Kaiming},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year      = {2025},
  note      = {Oral},
  url       = {https://openreview.net/forum?id=uWj4s7rMnR}
}

Unweighted Layered Graph Traversal

Xingjian Bai, Christian Coester, Romain Cosson

alphabetical order

SODA 2025 Symposium on Discrete Algorithms

arXiv

@inproceedings{bai2025lgt,
  title     = {Unweighted Layered Graph Traversal},
  author    = {Bai, Xingjian and Coester, Christian and Cosson, Romain},
  booktitle = {ACM-SIAM Symposium on Discrete Algorithms (SODA)},
  year      = {2025}
}

Fixed Point Diffusion Models

Xingjian Bai*, Luke Melas-Kyriazi*

* equal contribution

CVPR 2024 IEEE/CVF Conference on Computer Vision and Pattern Recognition

page arXiv code

@inproceedings{bai2024fixed,
  title     = {Fixed Point Diffusion Models},
  author    = {Bai, Xingjian and Melas-Kyriazi, Luke},
  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
  year      = {2024}
}

Goodhart's Law in Reinforcement Learning

Jacek Karwowski, Oliver Hayman, Xingjian Bai, Klaus Kiendlhofer, Charlie Griffin, Joar Skalse

ICLR 2024 International Conference on Learning Representations

arXiv

@inproceedings{karwowski2024goodharts,
  title     = {Goodhart's Law in Reinforcement Learning},
  author    = {Karwowski, Jacek and Hayman, Oliver and Bai, Xingjian and Kiendlhofer, Klaus and Griffin, Charlie and Skalse, Joar},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year      = {2024}
}

Sorting with Predictions

Xingjian Bai, Christian Coester

alphabetical order

NeurIPS 2023 Conference on Neural Information Processing Systems

arXiv poster code

@inproceedings{bai2023sorting,
  title     = {Sorting with Predictions},
  author    = {Bai, Xingjian and Coester, Christian},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year      = {2023}
}

Wasserstein Distributional Robustness of Neural Networks

Xingjian Bai, Guangyi He, Yifan Jiang, Jan Obloj

alphabetical order

NeurIPS 2023 Conference on Neural Information Processing Systems

arXiv poster code

@inproceedings{bai2023wasserstein,
  title     = {Wasserstein Distributional Robustness of Neural Networks},
  author    = {Bai, Xingjian and He, Guangyi and Jiang, Yifan and Obloj, Jan},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year      = {2023}
}

Chatting Time

Happy to chat about research, algorithms, or life in grad school. Pick a slot below, or email me directly at xbai@mit.edu.